Improved correlation between scores on the ... - Semantic Scholar

Brain (1998), 121, 1305–1312

Improved correlation between scores on the expanded disability status scale and cerebral lesion load in relapsing–remitting multiple sclerosis Results of the application of new imaging methods F. Riahi, A. Zijdenbos, S. Narayanan, D. Arnold, G. Francis, J. Antel and A. C. Evans Department of Neurology and Neurosurgery, Montre´al Neurological Institute and Hospital, McGill University, Canada

Correspondence to: Dr A. C. Evans, Montre´al Neurological Institute, 3801 University Street, Montre´al, Que´bec, Canada H3A 2B4

Summary We hypothesized that a better correlation between MRI and clinical measures of neurological disability using the expanded disability status scale (EDSS) in multiple sclerosis could be obtained by assessing lesion load only in and around the corticospinal tracts, since the EDSS is weighted towards motor and ambulatory deficits. Multiple sclerosis lesions in cerebral MRIs from 39 patients with relapsing–remitting multiple sclerosis were manually painted using a three-dimensional computer display tool and mapped into a standardized three-dimensional coordinate space. Total lesion load was then measured. A mask to expose only the corticospinal tract was extracted from an MRI atlas and used to measure lesion load in the corticospinal tract. To account for the residual anatomical variability among the different MRI volumes after stereotaxic transformation, the corticospinal tract mask was dilated to various degrees and the lesion load remeasured. Spearman’s rank correlation coefficient was

used to calculate the correlation between the EDSS and total lesion load and corticospinal tract lesion load and between the EDSS subscores and total lesion load and corticospinal tract lesion load. Spearman’s rank correlation coefficient between the EDSS and total lesion load was 0.6, probably reflecting the rather broad EDSS range represented in the study. The highest correlation of 0.67 was between the EDSS and corticospinal tract lesion load, dilated with a blurring kernel of 8–10 mm. The pyramidal subscore alone showed a weaker correlation with total lesion load, and with corticospinal tract lesion load, than did the overall EDSS, possibly reflecting the narrow range of disability in these subscores in patients with EDSS scores of 1–6.5. The imperfect correlation between the EDSS and corticospinal tract lesion load suggests that factors other than cerebral T2weighted lesion volume are important determinants of disability.

Keywords: multiple sclerosis; MRI; disability; lesion load Abbreviations: EDSS 5 expanded disability status scale; FWHM 5 full-width half-maximum; TE 5 echo time; TR 5 repetition time

Introduction The role of MRI in studying multiple sclerosis has been a subject of much debate (Frank et al., 1994; McDonald et al., 1994; Miller, 1994; Paty et al., 1994; Filippi et al., 1995a; Miller et al., 1996; Barkhof et al., 1997; Evans et al., 1997), particularly because of an imperfect correlation between clinical neurological disability and extent of MRI-defined disease. This lack of correlation could reflect variables derived from both the clinical and imaging measures used. The most frequently used and most standardized clinical scale of disability, the expanded disability status scale (EDSS) (Kurtzke, 1983), presents some problems in terms of © Oxford University Press 1998

subjective components (e.g. in the scoring of bowel and bladder symptoms) and intra- and inter-rater variability (Noseworthy et al., 1990), and is heavily weighted towards motor disturbances and ambulatory deficits (Willoughby and Paty, 1988). MRI assessment of disease, especially if performed using automated computer algorithms, is more objective and reproducible, and is not biased towards any particular clinical deficit. MRI measures of disease activity have been used as important secondary outcome measures in definitive clinical trials in the relapsing form of multiple sclerosis and as a primary outcome in pilot drug trials. Also,

1306

F. Riahi et al.

MRI activity measures reduce the size and length of clinical trials (Thompson et al., 1992; Nauta et al., 1994). A remaining challenge is to define optimal measures to demonstrate that a therapy reduces progression of disease and how MRI could serve as a surrogate marker for this. A number of studies have attempted to correlate the EDSS and MRI-defined lesion load, using a variety of techniques to measure the latter. Most have reported weak correlations, including Paty and Li (1993) (r 5 0.23 upon entry and 0.26 upon follow-up), van Walderveen et al. (1995) (r 5 0.30) and Gass et al. (1994) (r 5 0.33). Filippi et al. (1994) reported a correlation of 0.57 using a cohort of size and disability similar to that used in the current study. A number of reasons for this poor correlation have been suggested (McDonald et al., 1994; Miller, 1994; Filippi et al., 1995a) including insufficient follow-up time, methodological problems in assessment of lesion load, variety of subtypes of multiple sclerosis in the studied population, lack of imaging of spinal lesions, and pathological heterogeneity of the lesions seen on proton-density- and T2-weighted MRIs. We hypothesized that the correlation between the EDSS and corticospinal tract lesion load would be better than the correlation between the EDSS and total cerebral lesion load, given that the EDSS is weighted towards motor and ambulatory disturbances, that lesions in the corticospinal tract can cause motor disturbances, and that lesions in and around the corticospinal tract have been reported to be more clinically ‘eloquent’ than lesions elsewhere in the cerebrum (Zhao et al., 1993).

Method In this study, assessment of total and corticospinal lesion load from MRIs was performed automatically using the principles of stereotaxic analysis (Talairach and Tournoux, 1988; Evans et al., 1992; Evans et al., 1994a; Collins et al., 1995). The basis of the stereotaxic analysis is a threedimensional space (grid spacing 1 mm3) within which all brains have the same location, size and orientation. Registration is the process whereby two MRI datasets are aligned to establish a spatial correspondence between the corresponding morphological features. Each MRI is aligned with stereotaxic space by being registered automatically to an average (n . 300) MRI brain image volume aligned with stereotaxic space (Evans et al., 1994a). Registration to stereotaxic space has the advantage of removing gross variations in size among individual brains, allowing for the calculation of a normalized total lesion load. A further advantage of working in stereotaxic space is that the location of anatomical regions of interest such as the corticospinal tract can be obtained from a standardized brain atlas defined in stereotaxic space. A three-dimensional binary mask of any region of interest can be extracted from the atlas, which can then be applied to other MRI datasets in stereotaxic space to assess the lesion load within that region.

Table 1 Numbers of patients with each combination of EDSS and pyramidal subscore* Pyramidal subscore

EDSS 1

0 1 2 3

2 3 0 0

2

3

4

5

6

0 9 4 0

1 3 3 1

0 0 5 1

0 0 2 1

0 0 1 3

*For the 39 patients in this study.

Patient selection Cerebral MRIs from 27 patients with relapsing–remitting multiple sclerosis were chosen at random from a cohort of patients being followed as part of a clinical trial (the mean duration of disease was 7.0 years). The EDSS scores of these patients had been assessed by staff neurologists with no knowledge of the findings on MRI, and ranged from 1 to 4.5. In order to augment the range of EDSS scores, MRIs from another cohort of 12 patients with more advanced relapsing–remitting multiple sclerosis were also included. These represented all the patients with relapsing–remitting multiple sclerosis in a cohort previously studied. The EDSS scores in this group ranged from 3.0 to 6.5 and the mean disease duration was 9.6 years. [Although the mean disease duration in this group is greater than in the first group (of 27 patients), as expected given the higher average disability, the difference is not statistically significant; P 5 0.20.] Functional subscores were also obtained for pyramidal, cerebellar and sensory measures of disability to compare the correlation between lesion load and functional subscores related to motor and ambulatory function versus a functional subscore not directly related to these functions. Table 1 shows the relationship between the EDSS and pyramidal subscores. The relationship between the EDSS and cerebellar and sensory subscores is similar, with the patients with the high EDSSs having higher functional subscores.

MRI acquisition MRIs were acquired on a 1.5-T Philips Gyroscan or ACS III. MRIs for the 27 patients had been acquired as follows. Proton-density- and T2-weighted images were obtained with a two-dimensional double-spin echo with flow compensation, with repetition time (TR) 5 3000 ms, echo time (TE) 5 30, 80 ms, 3-mm contiguous transverse slices and a 256 3 192 mm matrix. The 12 other patients had T2 and proton-density imaging using a double-spin echo with TR 5 2116 ms, TE 5 30, 78 ms, 5.5-mm transverse slices with 0.5-mm interslice gap and a 256 3 256 mm matrix. For each patient and for each acquisition, the slices were reconstructed to generate a three-dimensional MRI volume.

Correlation of EDSS and lesion load in RRMS

Lesion labelling Lesion labelling was performed in real time under computermouse control using an interactive image-display tool that updates transverse, sagittal and coronal slices, and allows the user to toggle between the T2 and proton-density volumes. Lesions were identified and manually painted by one of us (F.R.). This process generated a total lesion mask for each patient: a volume with only lesion-containing voxels. The assessment of each volume in all three planes simultaneously, and in real time, allowed for labelling that was continuous in three-dimensions. This is in contrast to most previous studies where multiple two-dimensional slices were individually labelled, a process which can lead to inconsistencies in the third dimension.

Creation of the corticospinal tract mask A binary mask of the cerebral corticospinal tract was extracted, smoothed and resampled to the 1-mm3 resolution of stereotaxic space. Figure 1A shows the corticospinal tract mask overlaid on the average brain image in stereotaxic space. The mask was created using a brain atlas originally provided by Drs M. Shenton and R. Kikinis of Brigham and Women’s Hospital, Boston, Mass., USA, and subsequently registered onto stereotaxic space.

Creation of dilated corticospinal tract masks The mapping of each patient MRI volume into stereotaxic space was performed using a nine-parameter linear transformation. While this corrects for any linear variability between individual brains, there remains a residual non-linear anatomical variability of the order of 5–10 mm (Evans et al., 1991; Collins et al., 1994). Consequently, the corticospinal mask from the brain atlas will be a close, but not exact, fit for the corticospinal tract in each brain. To account for this residual non-linear anatomical variability, a dilating function was applied to the corticospinal tract mask. Initially, a threedimensional blurring function was applied to the binary corticospinal tract. The blurred mask was then converted back to a binary image. This resulted in a binary mask which was similar in shape to the original, but somewhat larger in all three dimensions and with smoother borders (see Fig. 1B). The amount of dilation is determined by the size of the neighbourhood over which the averaging takes place, and this is represented by the full-width half-maximum (FWHM) kernel. The corticospinal tract mask was dilated with FWHM kernels of 2, 4, 6, 8, 10, 20, 32 and 64 mm (the kernel of 64 mm creates a mask that encompasses almost all of the brain).

Measurement of total lesion load For each patient, the MRI datasets and total lesion volume were automatically registered into stereotaxic space according

1307

to the method of Collins et al. (1994), and resampled at the 1-mm3 resolution. Using an automated voxel-counting program the total lesion load was measured as a volume for each patient.

Measurement of the corticospinal tract lesion load For each patient, a corticospinal tract lesion volume was generated by voxel-wise multiplication of the total lesion volume by the corticospinal tract mask, thus leaving only those lesion voxels which are located in the corticospinal tract. Measurement of lesion load in each of the dilated corticospinal tract masks was done in an analogous manner.

Calculation of the correlation coefficient between the EDSS and lesion load The Spearman’s rank correlation coefficient was used to calculate the correlations between the EDSS and total lesion load, the EDSS and the corticospinal tract lesion load, and the EDSS and dilated corticospinal tract lesion load in stereotaxic space. Student’s t test was used to determine if the Spearman’s rank correlation coefficient was significantly different from zero for each of the above cases.

Calculation of the correlation between the EDSS subscores and lesion load To evaluate the relative contributions of EDSS subscores related to motor and ambulatory deficits and those less related to such deficits to the correlations calculated above, pyramidal, cerebellar and sensory EDSS subscores were extracted from the patients’ charts. The correlations between these subscores and the total lesion load, corticospinal tract lesion-load and dilated corticospinal tract lesion-load were measured as above.

Results Table 2 lists the Spearman’s rank correlation coefficients for the correlation between the EDSS and the total lesion load, the EDSS and the corticospinal tract lesion-load, and the EDSS and the dilated corticospinal tract lesion-load for FWHM kernels of 2–64 mm. With regard to the EDSS, the correlation is improved using the dilated corticospinal tract lesion-load with a maximum at 8–10 mm. Wider dilation begins to return the correlation to that of the total lesion load. Figure 2 illustrates these results in graphic form and again shows the best Spearman’s rank correlation coefficient at kernels of 8 and 10 mm. Table 3 lists the Spearman’s rank correlation coefficients for the correlation between the EDSS subscores and the total lesion load, corticospinal tract lesion-load and dilated corticospinal tract lesion-load (FWHM kernel 8).

1308

F. Riahi et al.

The pyramidal subscore shows a weaker correlation with all measures of lesion volume than the EDSS. As shown in Table 1, the pyramidal subscores in the current cohort covered only a narrow range (due to the ‘ceiling’ effect of functional scores) compared with the EDSS. This narrow range of disability would have an impact on the observed correlation. The cerebellar subscore also shows some correlation with the total lesion load and a better correlation with the dilated corticospinal tract lesion-load than the corticospinal tract

lesion-load. Sensory subscores showed no correlation with any MRI measures.

Discussion This study shows a correlation between the EDSS and total cerebral T2-weighted lesion load of 0.60. Although this is higher than in some previous studies which reported either no significant correlation (Koopmans et al., 1989; Thompson

Fig. 1 Transverse, sagittal, and coronal views of a corticospinal tract mask on the average brain volume (A), and of a dilated corticospinal tract mask (blurring kernel of 8 mm) on the average brain volume (B).

Correlation of EDSS and lesion load in RRMS et al., 1990), or significant correlation coefficients ranging from 0.2 to 0.3 (Paty and Li, 1993; Filippi et al., 1995b; van Walderveen et al., 1995), the value is similar to that found Table 2 EDSS and corticospinal tract lesion load correlations as functions of filter width* Filter width: FWHM (mm)

SRCC

t

0 2 4 6 8 10 20 32 64

0.57 0.64 0.66 0.66 0.67 0.67 0.62 0.6 0.6

4.27 5.03 5.31 5.42 5.44 5.44 4.83 4.59 4.59

*Spearman’s rank correlation coefficient (SRCC) and Student’s t at different levels of dilation. The equivalent value of unmasked tract lesion load was SRCC 5 0.6, t 5 4.59. All t-values are significant at P , 0.05.

by Filippi et al. (1994) using a cohort similar to the current one. We attribute the relatively strong correlation in this study to the wide range of EDSS scores in the patients studied. Registration of all the datasets onto stereotaxic space simplified the automatic measurement of lesion load within the corticospinal tract. Although the EDSS–corticospinal tract lesion-load correlation was slightly less than the EDSS–total lesion load correlation (0.57 versus 0.60), the correlation increased with the use of dilated corticospinal tract masks, reaching a peak of 0.67 at FWHM kernels of 8 and 10 mm before falling back to the EDSS–total lesion load correlation value of 0.60, as the increasingly dilated corticospinal tract masks eventually encompassed more of the brain (Fig. 2). The rationale for using dilated corticospinal tract masks was to compensate for the residual non-linear anatomical variability that remains after the stereotaxic transformation. Evans et al. (1991) and Collins et al. (1994) have estimated this residual non-linear anatomical variability at 6–7 mm of root mean square distance. (Conceptually, the root mean square distance approximates the radius of a sphere that would enclose 68% of all homologous landmarks.) This

Fig. 2 Graph of size of FWHM kernel in mm versus Spearman’s rank correlation coefficient (SRCC) for the correlation between the EDSS and corticospinal tract lesion load. Note that a kernel of 0 corresponds to the undilated corticospinal tract; total lesion load corresponds to a kernel of infinity (in practice, .64 mm).

Table 3 Spearman’s rank correlation coefficients for various scores and lesion loads

Total lesion load Corticospinal tract lesion load Dilated corticospinal tract lesion load with FWHM 5 8 mm

1309

EDSS

Pyramidal subscore

Cerebellar subscore

Sensory subscore

0.6 0.57 0.67

0.47 0.34 0.50

0.48 0.42 0.52

20.09 20.04 20.03

1310

F. Riahi et al.

variability is aggravated by the ventricular dilatation often seen in patients with advanced multiple sclerosis. Ideally, a full three-dimensional non-linear deformation would be applied to bring all brains into exact register, but this process is extremely computer-intensive and was not attempted for this study. Dilating the corticospinal tract mask is a simpler procedure to compensate for this residual non-linear mismatch in anatomy. Evans et al. (1994b) calculated that non-linear anatomical variability of 6–7 mm can be compensated for by using a blurring function with FWHM kernel of 8.2–9.4 mm. This corresponds well to our empiric finding that a FWHM kernel of 8–10 mm provides the best correlation between the EDSS and corticospinal tract lesionload. These findings support the hypothesis that the EDSS– corticospinal tract lesion-load correlation is better than the EDSS–total lesion-load correlation because of the weighting of the EDSS towards motor deficits. However, the maximum Spearman’s rank correlation coefficient of 0.67 is still far from perfect. The relatively small improvement in correlation between the EDSS and corticospinal tract lesion load versus total lesion load underscores the importance of other clinical and imaging factors in determining disability. Factors likely to be important include spinal cord lesions, pathological heterogeneity of the T2-weighted lesions seen on MRI, and abnormalities in white matter. The spinal cord is frequently involved in multiple sclerosis, and due to the concentration of descending motor tracts, even a small lesion in the cord can cause significant motor disability, which would increase the EDSS and reduce the correlation of the EDSS with cerebral lesion load. Losseff et al. (1996) have shown that clinical disability correlates with spinal cord atrophy, while Lycklama a Nijeholt et al. (1997) found that the presence of diffuse MRI signal abnormality in the spinal cord was associated with greater disability and a progressive clinical course. The pathological heterogeneity of multiple sclerosis lesions seen on MRI has been noted by Barnes et al. (1991) and McDonald et al. (1992). They found that focal increases in blood–brain barrier permeability, vasogenic oedema, demyelination, and axonal loss can all occur in the evolution of an multiple sclerosis lesion. These lesions all appear similar on unenhanced protondensity- and T2-weighted MRI, but are dissimilar in their reversibility and potential for causing disability, thus decreasing the correlation between imaging and clinical measures of disease. Proton MRI spectroscopy can provide more specific pathological information, particularly on the presence of axonal damage, not only in lesions but also in normal-appearing white matter (Arnold et al., 1994). Our hypothesis that the EDSS correlates better with the dilated corticospinal tract lesion-load than with the total lesion load might predict that an even stronger correlation would be found with a more specific measure of motor dysfunction, such as the pyramidal subscore of the EDSS. However, we did not find this. In our patient cohort, the range of pyramidal scores was narrow, which limited the usefulness of this index for correlative studies. Alternatively,

the pyramidal subscore may not be as good a measure of disability from corticospinal tract lesions as the EDSS. We also found a significant correlation between the cerebellar subscore and both total and corticospinal tract lesion load. This correlation could derive from the inclusion in the corticospinal tract masks of cerebellar afferent and efferent pathways, or from corticospinal dysfunction affecting tests of cerebellar function. Alternatively, lesions affecting the cerebellum and its connections could correlate highly with the corticospinal tract lesion-load, or disability from cerebellar lesions could correlate with disability from corticospinal tract lesions. One limitation of this study is the use of MRI with thicker (5.5-mm) slices for the 12 patients with higher EDSS scores. Recently, Filippi et al. (1997) have shown that measuring lesion load with 5-mm slices underestimates lesion load by 9% on average, compared with measuring lesion load with 3-mm slices. In this study the lesion load of the 12 patients with higher EDSS scores is underestimated by ~9%, which would therefore tend to reduce rather than increase any correlation between the EDSS and lesion load. In order to assess the effect that the different slice thickness may have had on our results, we adjusted the calculated lesion loads for the 12 patients who had been scanned at 5.5 mm upwards by 9%. The resulting correlations were essentially identical to our initial results: 0.61 versus 0.60 for the EDSS–total lesion load, 0.58 versus 0.57 for the EDSS–corticospinal tract lesion-load, and 0.67 versus 0.67 for the EDSS–dilated corticospinal tract lesion-load (FWHM 8 mm). Another limitation of studies of this type is the small cohort size, which can cause the correlations obtained to be influenced by a relatively small number of atypical observations. In our study, for the 27 patients scanned with 3 mm slices, the correlations between EDDS and total lesion load, corticospinal tract lesion-load and dilated corticospinal tract lesion-load (FWHM kernel of 8 mm) were 0.28, 0.34 and 0.35, respectively, while smaller cohorts of 12 and 15 patients showed even smaller correlation coefficients. We will address the issue of the relationship between cohort size and the EDSS–lesion load correlation in a future study. The use of stereotaxic space is advantageous for studies of this type; it permits the study of specific regions of the brain automatically by the use of masks, it corrects for the linear differences between individual brains and, through the use of dilated masks, accounts for the non-linear morphometric variations between individual brains. Future advances such as non-linear registration and fully automated algorithms for lesion volume measurement (Dawant and Zijdenbos, 1994) should make it possible to perform studies of this type on much larger numbers of patients with greater anatomical accuracy.

Conclusions The EDSS correlates better with corticospinal tract lesion load than with total lesion load once anatomical variability has

Correlation of EDSS and lesion load in RRMS been accounted for. However, this improvement is relatively small, emphasizing the importance of other clinical and imaging factors. Studies would benefit from better measures of pyramidal function and inclusion of MRI measures which evaluate spinal cord motor pathways, define the functional significance of T2-weighted multiple sclerosis lesions and assess the disease process in normal-appearing white matter.

References Arnold DL, Riess GT, Mattews PM, Francis GS, Collins DL, Wolfson C, et al. Use of proton magnetic resonance spectroscopy for monitoring disease progression in multiple sclerosis. Ann Neurol 1994; 36: 76–82. Barkhof F, Filippi M, Miller DH, Tofts P, Kappos L, Thompson AJ. Strategies for optimizing MRI techniques aimed at monitoring disease activity in multiple sclerosis treatment. J Neurol 1997; 244: 76–84.

1311

AJ, et al. Correlations between changes in disability and T2weighted brain MRI activity in multiple sclerosis: a follow-up study. Neurology 1995b; 45: 255–60. Filippi M, van Waesberghe JH, Horsfield MA, Bressi S, Gasperini C, Yousry TA, et al. Interscanner variation in brain MRI lesion load measurements in MS: implications for clinical trials. Neurology 1997; 49: 371–7. Frank JA, Stone LA, Smith ME, Albert PS, Maloni H, McFarland HF. Serial contrast-enhanced magnetic resonance imaging in patients with early relapsing-remitting multiple sclerosis: implications for treatment trials. Ann Neurol 1994; 36 Suppl: S86–90. Gass A, Barker GJ, Kidd D, Thorpe JW, MacManus D, Brennan A, et al. Correlation of magnetization transfer ratio with clinical disability in multiple sclerosis. Ann Neurol 1994; 36: 62–7. Koopmans RA, Li DKB, Grochowski E, Cutler PJ, Paty DW. Benign versus chronic progressive multiple sclerosis: magnetic resonance imaging features. Ann Neurol 1989; 25: 74–81.

Barnes D, Munro PM, Youl BD, Prineas JW, McDonald WI. The longstanding MS lesion. A quantitative MRI and electron microscopic study. Brain 1991; 114: 1271–80.

Kurtzke JF. Rating neurologic impairment in multiple sclerosis: An expanded disability status scale (EDSS). Neurology 1983; 33: 1444–52.

Collins DL, Neelin P, Peters TM, Evans AC. Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space. J. Comput Assist Tomogr 1994; 18: 192 – 205.

Losseff NA, Webb SL, O’Riordan JI, Page R, Wang L, Barker GJ. Spinal cord atrophy and disability in multiple sclerosis. A new reproducible and sensitive MRI method with potential to monitor disease progression. Brain 1996; 119: 701–8.

Collins DL, Holmes CJ, Peters TM, Evans AC. Automatic 3d model-based neuroanatomical segmentation. Hum Brain Mapp 1995; 3: 190–208. Evans AC, Dai W, Collins L, Neelin P, Marrett S. Warping of a computerized 3d atlas to match brain image volumes for quantitative neuroanatomical and functional analysis. Proceedings of SPIE— The International Society for Opitical Engineering; Vol. 1445 Image Processing 1991: 236–46. Evans AC, Collins DL, Milner B. An MRI-based stereotactic atlas from 250 young normal subjects [abstract]. Soc Neurosci Abstr 1992; 18: 408. Evans AC, Kamber M, Collins DL, MacDonald D. An MRI-based probabilistic atlas of neuroanatomy. In: Shorvon SD, Fish DR, Andermann F. Magnetic resonance scanning and epilepsy. Bydder GM, Stefan H, editors. New York: Plenum Press; 1994a. p. 263–74. Evans AC, Collins DL, Mills SR, Brown ED, Peters TM. 3D statistical neuroanatomical models from 305 MRI volumes. Proceedings of the IEEE Nuclear Science Symposium Medical Imaging Conference 1994b; 3: 1813–7. Evans AC, Frank JA, Antel J, Miller DH. Evaluation of multiple sclerosis lesion load: comparison of image processing techniques. Ann Neurol 1997; 41: 125–32. Filippi M, Horsfield MA, Morrissey SP, MacManus DG, Rudge P, McDonald WI, et al. Quantitative brain MRI lesion load predicts the course of clinically isolated syndromes suggestive of multiple sclerosis. Neurology 1994; 44: 635–41. Filippi M, Horsfield MA, Tofts PS, Barkhof F, Thompson AJ, Miller DH. Quantitative assessment of MRI lesion load in monitoring the evolution of multiple sclerosis. Brain 1995a; 118: 1601–12. Filippi M, Paty DW, Kappos L, Barkhof F, Compston DA, Thompson

Lycklama a Nijeholt GJ, Barkof F, Scheltens P, Castelijns JA, Ader H, van Waesberghe JH. MR of the spinal cord in multiple sclerosis: relation to clinical subtype and disability. Am J Neuroradiol 1997; 18: 1041–8. McDonald WI, Miller DH, Barnes D. The pathological evolution of multiple sclerosis. Neuropathol Appl Neurobiol 1992; 18: 319–34. McDonald WI, Miller DH, Thompson AJ. Are magnetic resonance findings predictive of clinical outcome in therapeutic trials in multiple sclerosis? The dilemma of interferon-beta. Ann Neurol 1994; 36: 14–8. Miller DH. Magnetic resonance in monitoring the treatment of multiple sclerosis. Ann Neurol 1994; 36 Suppl: S91–94. Miller DH, Albert PS, Barkhof F, Francis G, Frank JA, Hodgkinson S, et al. Guidelines for the use of magnetic resonance techniques in monitoring the treatment of multiple sclerosis. Ann Neurol 1996; 39: 6–16. Nauta JJ, Thompson AJ, Barkhof F, Miller DH. Magnetic resonance imaging in monitoring the treatment of multiple sclerosis patients: statistical power of parallel-groups and crossover designs. J Neurol Sci 1994; 122: 6–14. Noseworthy JH, Vandervoort MK, Wong CJ, Ebers GC. Interrater variability with the expanded disability status scale (EDSS) and functional systems (FS) in a multiple sclerosis clinical trial. Neurology 1990; 40: 971–5. Paty DW, Li DK. Interferon beta-1b is effective in relapsingremitting multiple sclerosis. II. MRI analysis results of a multicenter, randomized, double-blind, placebo-controlled trial [see comments]. Neurology 1993; 43: 662–7. Comment in: Neurology 1993; 43: 641–3.

1312

F. Riahi et al.

Paty DW, Li DK, Oger JJ-F, Kastrukoff L, Koopmans R, Tanton E, et al. Magnetic resonance imaging in the evaluation of clinical trials in multiple sclerosis. Ann Neurol 1994; 36 Suppl: S95–6. Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain. Stuttgart: Thieme; 1988. Thompson AJ, Kermode AG, MacManus DG, Kendall BE, Kingsley DP, Moseley IF, et al. Patterns of disease activity in multiple sclerosis: clinical and magnetic resonance imaging study [see comments]. BMJ 1990; 300: 631–4. Comment in: BMJ 1990; 300: 1272. Thompson AJ, Miller D, Youl B, MacManus D, Moore S, Kingsley D, et al. Serial gadolinium-enhanced MRI in relapsing/remitting multiple sclerosis of varying disease duration. Neurology 1992; 42: 60–3. van Walderveen MA, Barkhof F, Hommes OR, Polman CH, Tobi

H, Frequin SJ, et al. Correlating MRI and clinical disease activity in multiple sclerosis: relevance of hypointense lesions on short-TR/ short-TE (T1-weighted) spin-echo images. Neurology 1995; 45: 1684–90. Willoughby EW, Paty DW. Scales for rating impairment in multiple sclerosis: a critique. Neurology 1988; 38: 1793–8. Zhao GJ, Koopmans RA, Li DKB, Redekop K, Morrison W, Nelson J, et al. Corticspinal tract lesions in multiple sclerosis: relationship between MRI activity and clinical course [abstract]. Neurology 1993; 43 (4 Suppl): A246. Zijdenbos AP, Dawant BM. Brain segmentation and white matter lesion detection in MR images. Crit Rev Biomed Eng 1994; 22:401–65 Received September 15, 1997. Revised December 31, 1997. Accepted March 23, 1998